from usal.es - BISITE

Information Fusion 23 (2015) 43–57
Contents lists available at ScienceDirect
Information Fusion
journal homepage: www.elsevier.com/locate/inffus
Multi-Agent Information Fusion System to manage data from a WSN
in a residential home
Sara Rodríguez a,⇑, Juan F. De Paz a, Gabriel Villarrubia a, Carolina Zato a, Javier Bajo b, Juan M. Corchado a
a
b
Departamento Informática y Automática, Universidad de Salamanca, Plaza de la Merced s/n, 37008 Salamanca, Spain
Departamento de Inteligencia Artificial, Facultad de Informática, Universidad Politécnica de Madrid, Campus Montegancedo, Boadilla del Monte, Madrid 28660, Spain
a r t i c l e
i n f o
Article history:
Available online 26 March 2014
Keywords:
Multi-Agent Systems
Wireless Sensor Networks
Information fusion
Ambient Intelligence
a b s t r a c t
With the increase of intelligent systems based on Multi-Agent Systems (MAS) and the use of Wireless
Sensor Networks (WSN) in context-aware scenarios, information fusion has become an essential part
of this kind of systems where the information is distributed among nodes or agents. This paper presents
a new MAS specially designed to manage data from WSNs, which was tested in a residential home for the
elderly. The proposed MAS architecture is based on virtual organizations, and incorporates social
behaviors to improve the information fusion processes. The data that the system manages and analyzes
correspond to the actual data of the activities of a resident. Data is collected as the information event
counts detected by the sensors in a specific time interval, typically one day. We have designed a system
that improves the quality of life of dependant people, especially elderly, by fusioning data obtained by
multiple sensors and information of their daily activities. The high development of systems that extract
and store information make essential to improve the mechanisms to deal with the avalanche of
context data. In our case, the MAS approach results appropriated because each agent can represent an
autonomous entity with different capabilities and offering different services but collaborating among
them. Several tests have been performed to evaluate this platform and preliminary results and the
conclusions are presented in this paper.
Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction
Currently, there is a growing need to find more effective ways to
provide social services and medical care to the growing number of
elderly people with some kind of dependency [1]. This problem has
become one of the great challenges for Europe and its scientific
community. Some developments such as the AAL (Ambient
Assisted Living), sponsored by the IST (Information Society
Technologies) under the Seventh Framework Programme of the
European Union (European 7th Framework Programme), have
focused on finding new ways to address this problem from a technological point of view. As mentioned in [2], Ambient Intelligence
is considered ‘‘a new way to interact between people and technology, where the latter adapts to individuals and their context and
contains a range of interactive devices capable of meeting the
demands and requirements of the users’’. Moreover, it is important
to remark that the mechanisms and technologies that form an AmI
⇑ Corresponding author.
E-mail addresses: [email protected] (S. Rodríguez), [email protected] (J.F. De Paz),
[email protected] (G. Villarrubia), [email protected] (C. Zato), jbajo@fi.upm.es (J. Bajo),
[email protected] (J.M. Corchado).
http://dx.doi.org/10.1016/j.inffus.2014.03.003
1566-2535/Ó 2014 Elsevier B.V. All rights reserved.
system must allow their autonomous functioning without disturbing the people’s environment but making easy their daily activities.
Information fusion is understood as a process that gathers
assessments from the environment. It is based on goals and combines information at a low and high level. As a result, advantages
of intelligent approach as Multi-Agent Systems (MAS) within the
information fusion process have been recently emerging [3–5].
Agent Technology [6] is gaining progressively more importance
in the field of distributed and dynamic intelligent environments;
its participation in this process fulfils the requirements and goals
of the systems developed under the framework of the Ambient
Intelligence and Ambient Assisted Living. These intelligent systems
provide a powerful high-level tool and aim to support people in
several aspects of their daily life, this support includes the prediction of different dangerous situations, detecting physical problems
in people or buildings and also, provide a cognitive support. By
integrating intelligent and dynamic mechanisms to learn from past
experiences, the proposed architecture is able to provide users
with better tools for supplying healthcare.
Wireless Sensor Networks (WSNs) are mainly used to extract
information about the environment and behave consequently with
an interaction on it, extending users’ capabilities and automating
44
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
daily actions [7]. One of the most important applications for WSNs
is Real-Time Locating Systems (RTLS). As we previously studied in
[8], although outdoor locating is well covered by systems such as
the current GPS (Global Positioning System), indoor locating needs
still more development, especially with respect to accuracy and
low-cost and efficient infrastructures. The use of optimized
locating techniques allows obtaining more accurate locations using
even fewer sensors and with fewer computational requirements
[9].
As presented in [10] the innovative smaller, portable and
non-intrusive devices [11] are progressively more efficient when
gathering context-information [12]. Thus, the new Ambient Intelligence platforms should encourage the integration of such devices
in order to create open, flexible and adaptable systems. This reason
lead us to said that virtual organizations of agents are an ideal
option to create and develop the open and heterogeneous systems
such as those normally found in the information fusion process.
The system represented in this article arises from an increasing
need to research new solutions to meet the special needs of the
elderly, which will result in one of the most benefit segment of
population. The costs associated with these actions are no longer
bearable in terms of the investments in infrastructure and human
resources required for the traditional model (its survival is not
guaranteed). Innovative solutions are therefore needed in order
to better address the problem of aging.
More specifically, this paper presents an intelligent multi-agent
system aimed at improving healthcare and assistance to elderly
and dependent people in geriatric residences and at their homes.
The system is based on the PANGEA multi-agent architecture (Platform for Automatic coNstruction of orGanizations of intElligents
Agents) [13], which provides a high-level framework for intelligent
information fusion and management. The system makes use of
wireless sensor networks and a Real-Time Locating System to
obtain heterogeneous data, and provide autonomous responses
according to the environment status. The proposed system integrates a set of autonomous reactive and deliberative agents
designed to support the carergivers’ activities and to ensure that
the patients are given the proper care in both the residence and
their homes.
In summary, our work consists on developing a good tested system for information fusion technologies in an important real-world
application scenario, covering the following issues:
The development of a new Ambient Intelligence based multiagent system aimed at improving the healthcare of dependent
people in geriatric residences and in their own homes, focusing
on information fusion techniques and extending the system
proposed in [1].
The development under the PANGEA multi-agent architecture,
which provides a high-level framework for intelligent information fusion and management.
The appropriated use of virtual organizations of agents for the
overall management; and control systems for high-level sensor
data management.
To include a rule-based reasoning system to improve the accuracy of the results.
The use of virtual organizations of agents facilitate the incorporation of new information fusion techniques to the platform. Selfadaptive virtual organizations allow the dynamic incorporation of
specialized agents, which provides a framework for the incorporation of new information fusion techniques. The remainder of the
article is structured as follows: sections two and three review the
state of the art of related projects, both national and international,
as well as the various types of technology used in the study (information fusion, agents, rule-based reasoning systems, case based
reasoning), in an effort to identify current deficiencies. Section four
presents the proposed multiagent system, providing a general
description, its infrastructure and integration within sensor networks. This section specifically describes the basic structure of
the developed system, which is composed of three distinct parts:
real time identification and localization service [1]; telemonitoring
services; and an interface service for personnel. This article focuses
on telemonitoring services that permit carrying out a study, and
the observation and analysis of the users, which in this case are
the elderly living either in a care home or in their own home.
Finally, section five presents the results and conclusions obtained
from the study.
2. State of the art
The emergence of new technologies had resulted in a number of
projects that aim to improve interaction with an environment. One
specific application involves improving quality of life and care of
the elderly, as noted in the following specific projects:
The CommonWell European program [14] proposes an architecture to support citizens with limited mobility or an audio or
visual impairment. It focuses on the elderly but does not incorporate as many advanced adaptable or identification interfaces
or location.
The European DTV4A project proposes the use of digital television to integrate persons with limited abilities; however, the
television is the only mechanism to provide the services and
obtain information.
The European Monami project [15] proposes a global framework to offer services to the elderly and disabled. It focuses
on providing these services to individuals with the aim of
enabling a more independent lifestyle. It does not, however,
provide information for medical personnel or an alarm system.
At the national level we find the following:
The DISCATEL project [16] aims to facilitate contact with disabled people. It offers a monitoring system especially for people
with disabilities.
The INREDIS (Interfaces for Relations between the Environment
and people with disabilities) Project [17] is a CENIT project led
by Technosite, which combines the concept of using personal
devices with ubiquitous interoperability and characteristics to
strengthen accessibility for people with special needs.
The INCLUTEC group [18] is developing the eVia platform,
which is oriented toward the analysis and development of
new mechanisms to facilitate mobility such as advanced wheelchairs and specialized assistance vehicles, alternative and
improved communication, manipulation and cognition.
Although it can be applied to the elderly sector as a whole, this
project also focuses on persons with disabilities.
Apart from the projects in development, there are a high number of projects involving wearable health devices [19–21] integrated with sensors providing continuous monitoring of person’s
health related issues and daily activities. There are also many systems developed to monitoring and with the recovered information,
create alarms and facilitate the clinical decisions, i.e. Arezzo,
DeGeL, GLARE, GLEE, HeCaSe2 compared in [22]. SHAPHIRE [23]
is a system developed to provide clinical decision support for
remote monitoring of patients at their homes and at the hospital
to reduce the load of medical practitioners and healthcare costs.
The system CAMPH, a context-aware middleware for pervasive
homecare, is presented in [24]. The middleware offers several
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
key-enabling system services that consist of P2P-based context
query processing, context reasoning for activity recognition and
context-aware service management. However, camera based sensors for surveillance and security in which the images of the person
are taken require acceptability of the elderly which may not be
possible. BehaviorScope [25] processes streams of timestamped
sensor data along with prior context information to infer activities
and generate appropriate notifications. In this case, the activities of
interest are pre-programmed into a specification that is used by
the system to interpret the incoming sensor data stream. The system interprets the activities to generate summaries and other triggered notifications that are propagated to the users. However,
these projects are for very specific purposes.
There are also works centered on improving the activity patters
of the elderly, looking for a better learning of the software. This is
the case of the framework called DTFRA (Discovering of Temporal
Features and Relation of Activities), which focus on discovering and
representing the temporal features of activity patterns from sensor
data. The proposed algorithm is able to discover features and relations, such as the order of the activities, their usual start times and
durations by using rule mining and clustering techniques. Other
related works focus on this specific activity are [26–28]. In some
cases, these systems are extended with monitoring and modeling
the activities of daily living [29,30].
The most complete system found in our previous research was
JTH [31]. It is a prototype of wireless sensor networks, an interconnection platform and a service management platform to support
large scale data interconnections and real-time activity and health
state reports to related persons (e.g. doctors or nurses, elder-self,
relatives) via all popular communication approaches, such as automatic voice telephone call, SMS or Email etc. This system provides
4 main functionalities, including indoor monitoring, outdoor monitoring, activity and health state decision, emergency decision and
alarm.
The survey [32] about wireless sensor networks for healthcare
presents some interesting systems related to the field. Nevertheless, our system is technologically far away from those taken into
account. This system is more than a decision support system since
it integrates other capacities such as location, monitoring and prediction of dangerous situations. Moreover, it is based on the multiagent technology, which is highly appropriate to this end due to
the need of fusioning information from heterogeneous distributed
resources and autonomous entities. In this sense, the main contribution of this work is a system easily extensible that conforms a
complete tool for health-care services based on WSN sensors networks. The skeleton of the system is the PANGEA platform, which
provides the basic characteristics for the perfect functioning of the
agents.
What we are presenting is a system that aims to improve the
action of assistance in a senior home care facility. The system as
a whole offers real time identification and location services [1],
telemonitoring, and interface services for personnel. To this end,
techniques such as information fusion, rule-based reasoning systems, WSN and MAS are used. These techniques are presented
below.
2.1. Information fusion and agents
Most of the times, information fusion is a fundamental part of
sensor management. Waltz in [33] define sensor management as
‘‘any system which provides automatic control of a suite of sensors
or measurement devices and its. The main problem has to do with
the ability to observe a dynamic scene with a set of sensors by controlling their configuration, their sequencing or their state changes,
as well as the scheduling of the resources’’. Moreover, information
obtained from multiple sensors needs to be fused because no single
45
sensor can get all the information, and the information from different sensors may be uncertain, inaccurate, or even conflicting [34].
In our previous work [35] has been already exposed that there
is a considerable variety of sensors that can observe user contexts
and behaviors and multi-agent architectures that utilize data
merging to improve their output and efficiency. Such is the case
with Castanedo et al. [36], Pfeffer et al. [5], Liu et al. [37] or the system called HiLIFE [38].
Multiagent architectures are specially appropriated to implement the new algorithms for information fusion and to manage
high level information. This adequacy of agents and multi-agent
systems applied to information fusion has been deeper discussed
our previous work [39]. Open MAS [40] and virtual organizations
of agents [41–44,40,45], as a specialized version of MAS, are used
in this paper to allow the inclusion of organizational concepts
These concepts includes rules and norms [46], groups or institutions [47] and social structures [48].
From our perspective, there is already an open platform that has
been created and allows any type of configuration, adaptation
mechanisms, reorganization, search services, etc.
2.2. Rule-based reasoning systems
As observed in [49], the most critical step in information fusion
is related to the transformation from the observed parameters,
which are obtained by multiple sensors, and a decision or inference, which is calculated by fusion estimation and/or inference
processes. Finally, it is also important to get an easy understandable interpretation of the observed situations and the relationships
between them. In many cases, this interpretation needs not only
implicit information but also explicit data that must be extracted
via knowledge-based methods such as rule-based reasoning
systems.
The goal in building an automated reasoning system is to create
a system capable of making decisions based on the information
obtained through the heterogeneous sensors.
The developed system (and the agents supporting it) uses very
different techniques for base and meta-classifiers training as will
be seen in Section 4. They are used in the extraction of data and
the association rules from different sources of information. The
system applies information fusion algorithms which combine the
information gathered from each of the sensors through a casebased reasoning (CBR) mechanism [50]. But, in most cases, CBR
has not been used alone, but combined with various artificial intelligence techniques. Support Vector Machine (SVM) [51] has been
used with CBR in this study to perform the classification of the data
obtained by sensors and automatically create the intern structure
of the case base from existing data.
3. System overview
In this paper we propose a new Ambient Intelligence based
multi-agent system aimed at improving the healthcare of dependent people in geriatric residences and in their own homes, focusing on information fusion techniques and extending the system
proposed in [1]. The most important issue in this system is the
application of information fusion algorithms [52–54] that manage
data from Wireless Sensor Networks (WSN). The data collection is
carried out in real-time according to the action that occurs in the
environment, due to this, the agents can react accordingly in an
automatic and instantaneous way. Thanks to this configuration
the system enables the integration of an elevated number of WSNs
with the advantage of a greater simplicity due to the reuse of available resources.
46
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
The agents in the system are implemented within the agent
platform PANGEA (Platform for Automatic coNstruction of
orGanizations of intElligents Agents) [55,13]. PANGEA helps the
system to create agents in charge of all the functions of the system,
allowing them to organize and communicate more easily and
securely, regardless of how they are created, where they are
located, what data to collect or what role they play within the
system.
The two key concepts that come together in this system are: the
use of virtual organizations of agents for the overall management;
and control systems for high-level sensor data management. Furthermore, the sensors themselves will have to be managed and
analyzed to extract information from them and apply them to
the case study in question (healthcare and assistance to elderly
and dependent people in geriatric residences).
The context information includes not only information about
the environment, the people who lives in such environments is
also monitorized. This information includes parameters such as
location, temperature of the building, quality of the air, and heart
rhythm of the patient.
A node is each element that is included in a sensor network.
Each sensor node is usually formed by a microcontroller, a transceiver for radio or cable transmission, and a sensor or actuator
mechanism [11]. Some nodes act as routers, allowing them to forward data that must be delivered to other nodes in the network.
There are wireless technologies such as Wi-Fi, IEEE 802.15.4/ZigBee and Bluetooth that enable easier deployments than wired sensor networks [7]. At the level sensors, the basis of the WSN
infrastructure of the system is made up of several ZigBee nodes.
The ZigBee standard features make ZigBee an ideal supporting
wireless technology for building indoor Real-Time Locating Systems. The possibility of working with low-power nodes that do
not need large computational resources allows designers to reduce
hardware costs when implementing the systems. In addition, these
kinds of low-power nodes can reach a battery life of several years,
with regards to the transmission range (transmitted power), the
time resolution and the accuracy of the system. ZigBee-based
Real-Time Locating Systems can use different locating techniques
in order to estimate the positions of the tags in the environment.
In the proposed system, each ZigBee node includes an 8-bit RISC
(Atmel ATmega 1281) microcontroller with 8 KB of RAM, 4 KB of
EEPROM and 128 KB of Flash memory and an IEEE 802.15.4/ZigBee
transceiver (Atmel AT86RF230). These devices, called n-Core Sirius
B and Sirius D are shown in Fig. 1 [56].
At higher levels (features and decision), it is possible to detect
alterations in the environment and its corresponding response.
For example, if a change within a node (a change of light for
instance) is detected at the sensor level, the agents at a higher level
can decide to send a warning message or perform an action. The
actions and reactions are handled by the PANGEA platform since
all the agents that formed the organizations are designed with
the corresponding services and functions.
This configuration enables the capacity to manage a variety of
sensors, other devices of diagnostic and different sources of information (maintenance records, monitoring and observations). The
framework provides for information synchronization and highlevel fusion [57].
3.1. Infrastructure
The basic infrastructure of the system proposed in this article
consists of the following elements:
Wireless sensor infrastructure. This infrastructure is the foundation of the main system services. The infrastructure consists
of a set of physical wireless devices on which part of the low
level middleware will be executed and through which the rest
of the system can access its functionalities (for example, obtaining readings from various sensors or calculating the location of
the user).
Wireless devices. They form part of the wireless infrastructure
hardware. A set of low consumption and small wireless devices
were deployed. Each one shares a common basic architecture
composed of a microcontroller, a transceptor and a set of physical interfaces for the exchange of data between the device itself
and the sensors and actuators to which the device is connected.
According to the application, these devices can be either battery
or externally powered.
Low level middleware. A low level middleware is executed on the
wireless sensor network. It is composed of different layers that
permit the exchange of data between the wireless devices and
the other system components. The low level middleware contains a multiagent system [1,13,58] which permits the
exchange of information between the firmware included in
the wireless devices, the API communication to access the functionalities, and the communication protocol.
API communication. The API (Application Programming Interface) allows the rest of the system to interact with the wireless
devices in order to gather the information taken from the sensors they are connected to, and to send and receive data.
Agent-based infrastructure. This infrastructure, together with the
wireless sensor infrastructure, is the other pillar of the system.
It will be supported by PANGEA and will enable the transfer of
tasks that require a high computational cost and that, as a
result, are very difficult to implement in machines and devices,
often mobiles, used by the system clients. This infrastructure is
essentially composed of the agents that provide the system
functionality and that can be found in a remote server. The
agents execute the tasks (e.g., location). PANGEA also provides
the network communication data infrastructure, which makes
it possible to connect with client machines and for agents to
communicate with the services they offer.
Graphical interfaces. Last, but not least important, the system
includes a set of graphical interfaces that can display an
Fig. 1. n-Core Sirius devices.
47
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
Fig. 2. System architecture.
enriched form of all the information provided by the agents
based on the data they have received. This makes it possible
to access all the system information and the systems themselves using practically any device that can execute a simple
web browser (see Fig. 2).
All the artifacts, including sensors that make up the system
have been developed and combined with PANGEA platform to
obtain a system specifically oriented to manage information
obtained of the environment. For the communication between
agents, the IRC (Internet Relay Chat) Protocol is used in PANGEA
platform [55] demonstrating its robustness. IRC is a real time internet protocol for simultaneous text messaging or conferencing. IRC
provides advantages, such as ease of implementation and reliability, given that it has been widely used in online societies with good
functionality. It is used mainly for agent’s communication but also
allows private messaging for one on one communications, and
information transfers. All messages are the following format (see
Fig. 3):
The prefix is voluntary in some messages; the command is
one of the originals from the IRC standard. For example (see Fig. 4):
All agents that are created in PANGEA extend a common basic
structure that supports sending and receiving these messages. This
basic agent has implemented basic methods like ‘‘initialize’’ or
‘‘execute’’ which initializes the agent with some basic information,
or begin the implementation, respectively. A platform’s code
generating tool makes it possible to easily create an outline of
an agent, with the communication code requiring few lines of
code. The following lines of code are an example of this (see
Fig. 5).
The architecture was initially designed to be applied within
medical care environments, and defined three types of services
that can be easily adapted to other types of environment:
Real Time Identification and Location Service: system for locating elderly or dependant people and staff.
Telemonitoring service: system with different kind of sensors
that fusion the information obtaining knowledgment and alerts.
Caregiver support system: system that allows the caregivers to
set alarms and plan daily tasks and routines to take care of the
elderly.
Fig. 3. Message format.
Fig. 4. Example of message.
Fig. 5. Agent communication code.
The two first services are explained below in greater detail,
explaining how they are offered through the agents in the PANGEA
system.
3.2. Real Time Identification and Location Service
The Real Time Identification and Locating Service can identify
and know the position of the user or a particular object at any
given time. As with other services, it can also use algorithms to
manage alerts related to the location of the users and objects in
relation to the area in which they are found, permission, etc. It does
so by employing a set of location algorithms [52–54], which provides greater precision in locating persons than current location
systems. These algorithms fuse the information gathered from different sensors: ZigBee, Wi-Fi, accelerometers and compasses.
The configuration used in the system is similar to the configuration of our previous and well tested system presented in [1]. It consists of a ZigBee tag mounted on a bracelet worn on the users’ wrist
or ankle, several ZigBee readers installed throughout protected
zones, and a central workstation where all the information is processed and stored. These readers are installed all over the facilities
so that the system can detect when a user is trying to enter a forbidden area according to the user’s permissions profile. The ZigBee
48
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
network also allows obtaining information of the environment
from different sensors, such as temperature sensors, light sensors,
as well as smoke and gas detectors. In addition, different locating
techniques can be used as readers and tags carried by patients
and medical personnel. These devices are small enough to be carried by a patient, a caregiver or even an object, and have a battery
life of up to six months. The location of users is given as coordinated points obtained from the locating techniques provided by a
locating engine [59]. All information obtained by means of these
technologies is processed by the PANGEA agents. The system
allows users to keep track of any tag in the system as well as
receive distinct alerts in real-time coming from the system in
any Web-based device, such as PC or a smartphone carried by doctors and nurses. Some of the different alerts include panic button
alerts (when users press a panic button on their tag or in a fixed
device including such a button), forbidden area alerts (when users
enter a forbidden area according to their permissions), as well as
low-battery alerts (if a tag in the system should be recharged).
The ZigBee infrastructure was deployed in a 600 m2 area within
a residence housing dependent people with distinct types of
dementias such as Alzheimer’s disease. The locating infrastructure
is intended to provide the real-time position of people (i.e.,
patients and medical personnel) and assets (i.e., wheelchairs and
lifters) with an average accuracy of 2 m within the monitored area
in hallways and 0.65 m inside bedrooms, since these are the areas
which contain a greater number of sensors. Fig. 6 shows one of the
Web-based interfaces of the telemonitoring system in the monitored area.
In order to avoid problems produced by these algorithms, the
location system’s start-up procedure is modified. Instead of calculating the position of a tag based on the position of the reader, we
instead calculate a map of intensities for the environment. We take
the tag and calculate the RSSI levels obtained for each reader in the
different areas. Using this procedure, for every point (x, y) in a plan,
we obtain a set of measurements represented in:
1
1
n
n
mi ¼ ðxi ; yi ; node idi ; rssii ; . . . ; node idi ; rssii Þ
ð1Þ
where mi represents the measurement i, xi, yi represent the
j
x-coordinates and node idi , which is taken from the plans for
measurement, i represents the identifier for node j for measurement
j
i, and rssii represents the RSSI value of node j from measurement i.
S
Using m = mi we can build a classifier based on the data from m.
The classifier is incorporated into the system’s LocationAgent,
which is in charge of determining coordinate x according to the values from the RSSI signals obtained from the different readers. The
classes are defined according to the different pairs of values (xi, yi).
We have applied a Bayesian network. There are various Bayesian network search mechanisms, including tabu search [60], conditional independence [61], K2 [60], HillClimber [60], TAN (Tree
Argumented Naive Bayes) [62]. We have also used conditional
independence, an algorithm based on the calculation of the conditional Independence test for the variables to generate a DAG that
can obtain probability estimates. Assuming r pairs of different values for (xi, yi), the probability of measurement m belonging to class
i applying classifier C is defined as follows:
pi ¼ Cðm0 Þ
ð2Þ
where m is the class with k < r if
pk ¼ Maxðpi Þ with i ¼ 1 . . . r
ð3Þ
The algorithm simply places the tag in position (xi, yi) with
greater probability.
3.3. Telemonitoring service
The telemonitoring service can gather information from the
context and respond by generating alerts and other relevant
actions. To do so, it uses the contextual information taken from different sensors and from new knowledge extraction algorithms.
3.3.1. Rule-based telemonitoring
The telemonitoring service has the following objectives: (i)
monitor through the use of data gathered from sensors for each
participant; (ii) use the ‘‘Presence of Trace’’ of each person; (iii)
know the different available means of monitoring (devices, professional and support personnel, resources, etc.) and the communication that exists among them, and use the assistance device to
suggest certain actions to take; (iv) control and know what is
Fig. 6. Software interface of the locating system.
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
occurring in the global context of the residence of the person under
supervision; and (v) add and personalize rules as data are
expanded.
The service is offered through a remote monitoring center, a
sensor network, wireless actuators deployed within the environment, and a communication network that connects all the system
components.
As an example, we have a case study in which medical care is
provided in a care facility and private home. This requires an infrastructure of sensors and actuators deployed both in the care facility
and the home. The sensors gather information from the context,
that is, from both the elderly patient and the environment itself.
The information is processed and analyzed by a set of mechanisms
that facilitates the decision-making process and optimizes the
response of professional and support personnel. Those responsible
for assisting can access specific information regarding the elderly
patients: physical location in the home, medical history, high risk
situations detected, etc. Similarly, after providing assistance, the
data gathered from the actions taken are gathered for their subsequent analysis.
The information gathered by the infrastructure deployed in the
system is fundamental for the development of a ‘‘virtual assistance
services’’ system, which can be personalized for the home of each
person. The telemonitoring infrastructure is composed of a set of
Wireless ZigBee sensors [63] completely integrated within the
environment. The network in the system pilot was deployed in a
typical care facility bedroom (simulating the patient’s home) as
shown in Fig. 7. This space and the full residence were used to
carry out the tests.
Fig. 7 shows a map of the deployment of sensors in living quarters. In addition to the ZigBee readers (red squares), and sensors, 4
points (blue circles) were included in the movement pattern node:
one associated with the bathroom, one with the bed, one with the
dining table, and the other with the front door. The zig bee readers
must be located surrounding to the pattern nodes. The presence
sensors are located in positions avoiding blind spots. At this point,
it is important saying that here an example deployment is shown
as each person will have their individualized monitored deployment system and therefore its associated rules. One advantage is
49
that the system can adapt the system rules for each monitored person and the number and type of sensors associated. In the administration tool it is possible to select each of these monitored
persons to observe the distribution of deployed sensors. And the
rule system is customized through a case-based reasoning as it is
explained below.
The alerts generated by the sensors are managed through a
multiagent architecture that applies algorithms to manage the
information and make decisions. The alarm management system
is based on a behavioral module which applies the Drool production rule system, allowing the information to be processed quickly
and decidedly. The architecture applies information fusion algorithms [1] which combine the information gathered from each of
the sensors installed in the environment and the tracking information for each user to generate intelligent alerts through a casebased reasoning (CBR) mechanism [50]. The system includes a
‘‘daily activity planner’’ and a library of ‘‘rules’’ that generates
alerts when something occurs contrary to the typical daily plan.
For example: ‘‘If the subject remains in bed longer than eight
hours, an alert is generated’’.
The detection of anomalous behaviors is done through the
detection of strange movement patterns of the users. In order to
detect these patterns, a series of relevance variables are established to measure a series of variables in each of the nodes in the
graph that represent possible movement patterns. The values indicated in Table 1 are used for each node representing a movement
pattern. The nodes for each movement pattern do not include the
sample points in the map of intensities to reduce the number of
values in Table 1.
The agent that detects anomalous behaviors is based on the use
of predefined rules that are executed over the data from the previous table. An example of the rules can be seen in Table 1. The rules
are written in a text file and they can be modified without building
the source code. The Drools production rule system [64] was used
to configure the rules engine. Drools is a business rule management system (known as BRMS, Business Rule Management System) with a forward chaining inference based rules engine, using
an implementation oriented toward Rete algorithm objects [65].
In other words, it is a rules engine based on Java, responsible for
Fig. 7. Map of the deployment of sensors in the living quarters.
50
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
Table 1
Tracking data used. The variables used in Eq. (4) are represented in brackets.
Day of week (w)
Hour (h)
Node (n)
User (u)
Seconds (s)
Previous node (p)
Previous time (l)
Number of times (t)
Table 2
Sensor information. The variables used in Eq. (4) are represented in brackets.
1. . .7
Seconds starting at 0:00
Current node
User id
Seconds from the previous node
ID of previous node
Seconds since the node was last accessed
Number of times user has been in node since 0:00
applying business rules in our applications. This makes it possible
to abstract the code and process the objects and a relatively high
level with a language that is more simple and closer to natural language, separating rules from actions (see Fig. 8).
The Drool production rule system [64] is also used to manage
the information provided by the sensors. The information gathered
from the sensors makes it possible to know the state of the environment as well as the time instance in which the change was
made and the number of measurements taken in that particular
state, as shown in Table 2.
In addition to the previous rules, case based reasoning is used to
generate additional rules based on J48 to detect new cases identified as anomalous. If any of the procedures detects anomalous
behavior, an alarm will sound, unless a Drool rule to cancel an
alarm that has been activated through J48 is detected. The ability
to automatically cancel rules is to avoid launching a false positive
and allow the automatic cancellation of an alarm considered to be
incorrect.
The first step in defining the case-based reasoning is to establish
the information according to Tables 1 and 2, as shown in (4). The
variables are described in Tables 1 and 2. The case base is divided
into similar groups in order to improve the prediction capability of
the system.
ci ¼ ðwi ; hi ; ni ; ui ; si ; pi ; li ; ti ; Di Þ
[ i
Di ¼ dj
j
ð4Þ
i
dj ¼ ðidj ; dt j ; dsj ; dmj ; daj Þ
Once the concept of case has been defined, the CBR cycle is
defined for each of the following steps: retrieve, reuse, revise and
retain. As new cases are received in the system, they are introduced into the reasoning cycle. A prediction is made regarding
the need to generate an alarm. The complete process is described
in Fig. 9.
The retrieve phase is initiated upon receipt of a new case cn+1.
When the new case arrives, it is associated or classified with one
of the groups gk into which the memory is divided. Once the group
Sensor (id)
Time (dt)
State (ds)
Measurements (dm)
Alarm (da)
Id sensor
Date of last change
State of the sensor
Number of measurements taken in the state
Alarm activated
is determined, the J48 classifier associated with the set is recovered
or created, if it did not previously exist. In order to perform the
classification, the SVM [51] is applied, as part of the process indicated in Fig. 10a.
During the revise phase, the classifier retrieved in the previous
phase is applied to the new case, and a prediction is obtained.
During the retain phase Fig. 10b it is determined whether the
prediction made in the revise phase was satisfactory; if it was
not, the prediction is updated. During the revise phase the expert
confirms the prediction and during the retain phase the prediction
is stored according to a results obtained. That is, the system is
capable to detect new cases identified as anomalous, is capable
to automatically cancel rules is to avoid launching a false positive
of an alarm. In this sense, if in the revise phase the expert confirms
that the alarm was a false positive is stored. If the prediction is
incorrect, the new case is introduced into the system and the process of clustering and creating a classifier is carried out for each of
the clusters created. The EM method is used to create the clusters
since it facilitates the creation of groups without requiring the
number of clusters to be previously indicated. Additionally, it
works with nominal data, a process indicated. Finally, the SVM is
constructed according to the new groups and the system constructs the decision tree J48 for each group.
3.4. Integration within the multiagent system
PANGEA (Platform for Automatic coNstruction of orGanizations
of intElligents Agents) [13] is an agent platform to develop open
multi-agent systems that can manage roles, norms, organizations
and suborganizations, and facilitate the inclusion of organizational
aspects. The PANGEA platform has been adapted to be applied in
context-aware environments, with special attention placed on
obtaining new algorithms and modules for the fusion of contextual
information. This is a new perspective where social aspects are
taken into account to model and manage intelligent environments.
The basic agent types needed to the adequate functioning of the
platform and the main characteristics of it are defined in [55].
Four specialized suborganizations have been created in the
platform, all of which are managed by an OrganizationalAgent; this
helps the OrganizationManager to control the correct functioning.
Fig. 8. Example of rules.
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
51
Fig. 9. CBR reasoning cycle.
Fig. 10. (a) Algorithm retrieve and reuse phase, (b) revise and retain phase.
Each suborganization is specialized according to the tasks carried
out. The four suborganizations, which can be seen in Fig 5 are:
InformationProviderOrganization: enables interaction with the
user. As three different interfaces have been developed, there
is an agent for each one of them.
HomeAutomationOrganization: composed of all the agents that
control the different sensors of the spaces (presence, smoke,
temperature, etc.). All the information collected by them is sent
to the ZigbeeSupervisorAgent. After a previous evaluation, this
agent communicates the information to the InformationAgent.
The InformationAgent is the interface with the database.
LocatingOrganization: composed of all the agents that control
the Zigbee sensors and tags, which allows locating the people
(caregivers or patients). All the information collected by these
agents is sent to the ZigbeeSupervisorAgent. This information
is used by the MonitorAgent and the LocatingAgent if the user
wants to track a person.
CaregiverOrganization: each caregiver (doctor or nurse) is represented by an agent that collects information related to their
respective actions. Each action has a representative collection
of information that allows for knowledge extraction. This information is key for the proper operation of RulesAgent As it can,
for example, create the routine for each patient. When the information provided by the caregivers and sensors falls outside the
patient’s routine or indicates a dangerous situation, the AlarmAgent extends the alarm throughout the system. The MonitorAgent can receive the alarm because of its communication
with the SnifferAgent and shows it on the screen (see Fig. 11).
4. Results and conclusions
As indicated in the previous section, a case study was prepared
in a medical care environment for the elderly. The proposed system
was implemented in the ‘‘El Residencial La Vega’’ care facility in the
city of Salamanca and was tested over a period of 8 months. The El
52
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
Fig. 11. Overview of the deployed agents in PANGEA.
Residencial La Vega facility has residence areas for the elderly as
well as apartments similar to their patients’ home. We would like
to highlight the level of participation of the facility’s medical personnel in implementing the prototype.
In this results section we have included some of the most relevant
results obtained in the implementation of this system in the
residence. Specifically, in this section, several aspects are important:
– On one hand, the possible functions of the system, showing
some of the tools created (web and mobile tools) used by users
of the system and results show the fusion of sensor data and
information stored in the system in a distributed manner. All
these functionalities are implemented with PANGEA, so we
must make it clear that these features are just a sample of what
could turn out to be because this platform is very easy to scale
and the system can be enhance.
– On the other hand, the results of the rule system and the location algorithm used with the sensors installed in the residence
are shown. In this sense, favorable results are shown in a possible evolution of the system of rules.
First, we propose to study the reasons for residence and location
where they occurred (areas) taking as input data the observed data
from the sensors in different areas. In short, put in any space,
whether conceptual or geometric, the reasons and the situation
inside the residence, integrating them in context, and interpret that
situation. The residence is divided into zones (red, blue, yellow or
green). Each contains a type of residents. For example, the blue area
is indicated for patients with some form of dementia such as Alzheimer’s and green area for seniors who can fend for themselves.
About installation, the system services were put in place for
some elders, taking into account the considerations of the staff of
the residence. The growing use of telemonitoring to support independent living in a residence inevitably carries out many ethical
questions regarding privacy. A resident’s right to confidentiality
must be respected in any feature of healthcare, telemonitoring
included. At the resident side, the intrusiveness of the continuing
analysis of the resident in his/her own private life should be
minimized as much as possible. In this sense, approval for the
study was obtained from the Residencial La Vega Committee and
the University of Salamanca. Moreover, all personal data were
transformed previously to the storage experiment changing them
for general identifiers and data encryption and secure methods
were applied to ensure confidentiality of the data during transfer
over the system. At the monitoring end, access to data was
restricted using a hierarchical password system for the persons
implied in the experiments.
As a result of the information obtained from both the telemonitoring sensors and those of the caregivers and doctors, the system
agents were able to provide a system capable of interacting to control the state of the patient through a web tool.
The information obtained will permit the system to adapt
dynamically to the needs of the environment and of the patient.
For example, the second graph in Fig. 12 shows how a very different
number of assistive actions are taken in the different areas of the
residence (red, blue, yellow or green).1 The x-axis represents the
hours in a day and the y-axis the number of attendances by falls.
The green area underscores the number of falls between the hours
of 9 and 12 noon. If we take into account the fact that the green area
is where the residents with a specific degree of dependency are
located, we can adapt the number of assigned personnel to meet
these needs. Additionally, the responsible personnel can use mobile
devices to locate patients in real time, plan daily work tasks and
respond to alarms. Personnel in the residence can access details
regarding an alert and once it has been resolved, they can close the
alert with a description of what occurred and the reason for it. Likewise, it is possible to consult the different alerts made in each area
with respect to the time or the month they occurred, which is useful
information for creating patterns. The graphical representation of this
information can be done from the web page or from a mobile device.
The operator can also modify the access data and view a summary of
their activity. These three functions can be seen in Fig. 13.
The PANGEA multi-agent platform is used to implement the
system and the information fusion process. There are several
1
For interpretation of color in Fig. 12, the reader is referred to the web version of
this article.
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
53
Fig. 12. Web interfaces to query statistics. Top image: Number of calls in the residence by sections. Bottom image: Graph of falls (y axis) per hour (x axis).
reasons to prefer such a platform: a multi-agent system represents
an advantageous paradigm for the analysis, design and implementation of complex systems. An Information Fusion application is
naturally distributed: data sources are distributed; the users use
the results produced by the system are distributed, and data processing is performed in a distributed manner. PANGEA provides
the infrastructure to create our multi-agent system and also to
gather the data, communicate with users, and process data. If data
from different sources are private or classified, PANGEA makes it
possible to create different forms of access which are not available
for centralized processing. Additionally, the system can be large
scale in the future, and naturally deconstructed as well; this allows
it to takes the most advantage of the benefits provided by the MAS
(and PANGEA) approach.
To test the system’s performance, we analyzed the precision in
locating and detecting anomalous behaviors. The precision of the
location algorithm was analyzed by comparing the mean absolute
error obtained from applying the proposed procedure to the means
obtained from multilateration and signpost. Besides, the Bayesian
network was replaced by others classifiers in order to compare
the performance. The calculation in the maps of intensities was
performed by measuring the steps taken when walking through
the different hallways and rooms. In each monitored area, a reference point was used to represent the area, manage the information,
and therefore reduce the information that needed to be stored. In
each room, a mesh was used to simulate movement patterns and
obtain the surface of the room that was measured. The mean absolute error obtained is shown in Table 3.
In order to carry out the alarm detection procedure, the users’
movement patterns were stored over a period of one week. During
this time, the patterns were classified to include unusual behavior.
The number of measurements taken is 983, following the information shown in Tables 1 and 2. A total of 173 unusual situations
were identified. This information was used to create a decision tree
that was subsequently used to perform the classification. The rules
that were generated are similar in form to the following rule: if
User = 13 and Node! = 13,001 and sensor = 13,002 and state = 1
and measurements >10 then alarm = 1.
54
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
Fig. 13. System interfaces. (a) Description of an alert, (b) number of assistances by zone. Graph lines: the color of the line represents the zone, x axis the hour and y axis the
number of assistances during the hour of x axis. Bar graph: the number of assistance during 8 months (y axis) for each zone (x axis), and (c) the information of an user. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Table 3
Error comparison between methods applying maps of intensity and algorithms
without applying maps of intensities.
Model
Proposed model
Proposed model
Proposed model
Proposed model
Multilateration
Signpost
Mean absolute error
with
with
with
with
Bayesian network
J48
SVM
IBK
1.188
1.648
1.789
1.475
1.981
2.371
The rule refers to user 13, which is why the sensor identifiers
and the nodes in the user’s bedroom begin with 13. When 10 consecutive measurements are detected in a position different from
the closest node, which is located in the refrigerator, an alarm will
sound. When the rules are detected, as indicated in rule 1, where
all the sensors and the user belong to the same room, they are generalized for all users so that the user ID, sensor and node information are all modified for the other users.
The system was cross validated applying the one-leave-out
technique, which resulted in an average prediction rate of 93.5%.
The percentage of false negatives rose to 1.2% and the remainder
were false positives.
In order to analyze the evolution of the system, we began with
500 cases and introduced new cases into the system until reaching
983, as previously indicated. Fig. 14 shows the change in the number of groups in the case base, the evolution of the number of cases
and the average number of rules for each group. The diagonal
graphs represent the function of density for the indicated variable.
For the remaining cells, file i, column j, the x-axis represents the
variable indicated in cell ii, while in the y-axis the variable is associated with cell jj. The dot points represent each element; the red2
line represents the tendency according to the represented element,
the green line the regression rect. For example, in file 1, column 2
we can see the variation of the average number of rules as the
2
For interpretation of color in Fig. 14, the reader is referred to the web version of
this article.
55
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
600
700
800
900
1000
16.0
500
1000
14.0
14.5
15.0
15.5
average.rules
6.0
500
600
700
800
900
cases
4.0
4.5
5.0
5.5
groups
14.0
14.5
15.0
15.5
16.0
4.0
4.5
5.0
5.5
6.0
Fig. 14. Evolution of the system.
number of cases increases from 500 to 1000. We can see in the red
line that the tendency increases as the number of cases also
increases. However, we can see in the file 1 column 3 that the average number of rules for each group varies between 13 and 16. Similarly, the number of rules also increases, although to a lesser extent,
with the number of groups, as shown in the graph shows.
As we can see in the results, the system proposed in this article
exceeds the performance of the system currently used in the La
Vega Residence, and improves the existing teleassistance system,
providing learning and adaptation capabilities. These systems are
currently very limited, as they require the elderly patients to make
a conscientious and deliberate effort in their home (for example,
pressing the panic button) and do not offer an automated and
intelligent detection of high risk situations.
The system proposed in this article implements a significant
number of the concepts used in Ambient Intelligence, in an
attempt to reduce the interaction of the elderly patient as much
as possible and create a much more direct communication with
the professional and support staff responsible for providing assistance. The medical personnel at the residence noted the system’s
ease of use, as well as the support in providing patient care. The
aim was to facilitate the daily activities of the elderly patients in
a way that is both ubiquitous and intuitive, in addition to optimizing the assistance services and improving quality of life. The proposed system represents an advancement with regard to existing
platforms, as shown in its application in the senior care facility
case study. It was for this reason that the case study incorporated
a set of devices that comprise sensors, id or locating elements, push
button actuators and interactive elements such as screens. The
study also incorporated a control center that interacts continuously with the system and a data repository, which is very useful
for tracing services and the personalization of spaces and services.
Acknowledgements
This work has been carried out by the project Sociedades
Humano-Agente: Inmersion, Adaptacion y Simulacion. TIN201236586-C03-03. Ministerio de Economía y Competitividad. Fondos
Feder. Special thanks to Limcasa and Flag Solutions for their support, as well as the technology provided.
References
[1] R.S. Alonso, D.I. Tapia, G. Villarrubia, J.F. De Paz, Agent technology and wireless
sensor networks for monitoring patients in residences and their homes, in:
11th International Conference in Practical Applications of Agents And
Multiagent Systems – Workshop on Agents for AAL and E-health, Salamanca,
Spain, 2013, pp. 417–418.
[2] E. Aarts, R. Wichert, Ambient Intelligence Technology Guide, Springer Verlag,
Berlin, 2009.
56
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
[3] F. Castanedo, M.A. Patricio, J.M. García, J. Molina, Information fusion to
improve trajectory tracking in cooperative surveillance multi-agent
architecture, Inf. Fusion 11 (2010) 243–255.
[4] Y. Liu, S. Wang, X. Du, A multi-agent information fusion model for ship
collision avoidance, in: IEEE International Conference on Machine Learning and
Cybernetics, Kunming, China, 2008, pp. 6–11.
[5] A. Pfeffer, S. Das, D. Lawless, B. Ng, Factored reasoning for monitoring dynamic
team and goal formation, Inf. Fusion 10 (1) (2009) 99–106.
[6] M. Wooldridge, N.R. Jennings, Agent Theories, Architectures, and languages: a
survey, in: ECAI-94 Proceedings of the Workshop on Agent Theories,
Architectures, and Languages on Intelligent Agents, Amsterdam, The
Netherlands, 1995, pp. 1–22.
[7] J. Sarangapani, Wireless Ad hoc and Sensor Networks: Protocols, Performance,
and Control, first ed., CRC Press, Boca Raton, 2007.
[8] J.F. De Paz, D.I. Tapia, R.S. Alonso, C.I. Pinzón, J. Bajo, J.M. Corchado, Mitigation
of the ground reflection effect in real-time locating systems based on wireless
sensor networks by using artificial neural networks, J. Knowledge Inf. Syst. 34
(1) (2013) 193–217.
[9] C. Nerguizian, C. Despins, S. Affès, Indoor geolocation with received signal
strength fingerprinting technique and neural networks, in: 11th International
Conference on Telecommunications, Fortaleza, Brazil, 2004, pp. 866–875.
[10] S. Rodríguez, F. De la Prieta, E. García, C. Zato, J. Bajo, J.M. Corchado, Virtual
organizations in information fusion, in: 9th International Conference in
Practical Applications of Agents and Multiagent Systems – Special Session on
Adaptive Multiagent Systems, Salamanca, Spain, 2011, pp. 195–202.
[11] M. Marin-Perianu, N. Meratnia, P. Havinga, L. de Souza, J. Muller, P. Spiess, S.
Haller, T. Riedel, C. Decker, G. Stromberg, Decentralized enterprise systems: a
multiplatform wireless sensor network approach, IEEE Wirel. Commun. 14
(2007) 57–66.
[12] M.L. Borrajo, J.M. Corchado, E.S. Corchado, M.A. Pellicer, J. Bajo, Multi-agent
neural business control system, Inf. Sci. 180 (6) (2010) 911–927.
[13] C. Zato, G. Villarrubia, A. Sánchez, I. Barri, E. Rubión, A. Fernández, C. Rebate,
J.A. Cabo, T. Álamos, J. Sanz, J. Seco, J. Bajo, J.M. Corchado, PANGEA – Platform
for Automatic coNstruction of orGanizations of intElligent Agents, in: 9th
International Conference on Distributed Computing and Artificial Intelligence,
Salamanca, Spain, 2012, pp. 229–239.
[14] CommonWell Project <http://commonwell.eu/index.php> 2010.
[15] Monami project <http://www.monami.info/> (accessed December 2013).
[16] DISCATEL <http://www.aeerc.com/teletrabajo.cfm> (accessed December
2013).
[17] INREDIS <http://www.inredis.es/> (accessed December 2013).
[18] INCLUTEC <http://evia.imasdtic.es/es/Inicio/corporativo/grupos_de_trabajo/
inclutec/contenido.aspx> (accessed December 2013).
[19] H. Yu-Jin, K. Ig-Jae, C.A. Sang, K. Hyoung-Gon, Activity recognition using
wearable sensors for elder care, in: 2nd International Conference on Future
Generation Communication and Networking, Hainan Island, China, 2008, pp.
302–305.
[20] P. George, X. George, P. George, Monitoring and modeling simple everyday
activities of the elderly at home, in: 7th IEEE Consumer Communications and
Networking Conference, Las Vegas, USA, 2010, pp. 1–5.
[21] W.L. Seon, K. Yong-Joong. L. Gi-Sup, O.C. Byung, L. Nam-Ha, A remote
behavioral monitoring system for elders living alone, in: 7th International
Conference on Control, Automation and Systems, Seoul, Korea, 2007, pp.
2725–2730.
[22] D. Isern, A. Moreno, Computer-based execution of clinical guidelines: a review,
Int. J. Med. Informatics 77 (12) (2008) 787–808.
[23] L. GokceB, A.D. Asuman, A.O. Mehmet, A.T. Ibrahim, A.Y. Mustafa, A.O. Alper,
SAPHIRE: a multi-agent system for remote healthcare monitoring through
computerized clinical guidelines, in: R. Annicchiarico, U. Cortes, C. Udiales
(Eds.), Agent Technology and e-Health, Whitestein Series in Software Agent
Technologies and Autonomic Computing, Springer Verlag, Berlin, 2008, pp. 25–
44.
[24] K.P. Hung, G. Tao, X. Wenwei, P.P. Palmes, Z. Jian, N. Wen Long, W.T. Chee, H.C.
Nguyen, Context-aware middleware for pervasive elderly homecare, IEEE J.
Sel. Areas Commun. 27 (4) (2009) 510–524.
[25] D. Lymberopoulos, A. Bamis, T. Eixeira, A. Savvides, BehaviorScope: real-time
remote human monitoring using sensor networks, in: Proceedings of the
International Conference on Information Processing in Sensor, Networks, 2008,
pp. 533–534.
[26] P. Rashidi, D.J. Cook, Keeping the resident in the loop: Adapting the smart
home to the user, IEEE Trans. Syst. Man Cyber. 39 (5) (2009) 949–959.
[27] V.R. Jakkula, D.J. Cook, Using temporal relations in smart environment data for
activity prediction, in: Proceedings of the 24th International Conference on
Machine Learning, Corvallis, Oregon, 2007, pp. 1–4.
[28] B. Gottfried, H.W. Guesgen, S. Hubner, Spatiotemporal reasoning for smart
homes, in Designing Smart Homes, The Role of Artificial Intelligence, Springer,
2008, pp. 16–34.
[29] P. George, X. George, P. George, Monitoring and modeling simple everyday
activities of the elderly at home, in: Proceedings of the 7th IEEE Consumer
Communications and Networking Conference, 2010, pp. 1–5.
[30] H. Medjahed, D. Istrate, J. Boudy, B. Dorizzi, Human activities of daily living
recognition using fuzzy logic for elderly home monitoring, in: Proceedings of
the IEEE International Conference on Fuzzy Systems, 2009, pp. 2001–2006.
[31] H. Huo, Y. Xu, H. Yan, S. Mubeen, H. Zhang, An elderly health care system using
wireless sensor networks at home, in: Sensor Technologies and Applications
SENSORCOMM’09, 3rd International Conference, 2009, pp. 18–23.
[32] H. Alemdar, Cem Ersoy, Wireless sensor networks for healthcare: a survey,
Comput. Networks 54 (15) (2010) 2688–2710.
[33] E.L. Waltz, J. Llinas, Multisensor Information Fusion, Artech House, Norwood,
USA, 1990.
[34] H. Luo, S. Yang, X. Hu, X. Hu, Agent oriented intelligent fault diagnosis system
using evidence theory, Expert Syst. Appl. 39 (3) (2012) 2524–2531.
[35] D.I. Tapia, J.A. Fraile, A. De Luis, J. Bajo, Healthcare information fusion using
context-aware agents, in: Manuel Graña Romay, Emilio Corchado, and M.
Teresa Garcia Sebastian (Eds.), Proceedings of the 5th International Conference
on Hybrid Artificial Intelligence Systems – Volume Part I (HAIS’10), SpringerVerlag, Berlin, Heidelberg, 2010, pp. 96–103.
[36] F. Castanedo, J. García, A.M. Patricio, J.M. Molina, Information fusion to
improve trajectory tracking in a cooperative surveillance multi-agent
architecture, Inf. Fusion 11 (3) (2009) 243–255.
[37] Y.H. Liu, S.Z. Wang, X.M. Du, A multi-agent information fusion model for ship
collision avoidance, in: Proceedings of the International Conference on
Machine Learning and Cybernetics, 2008, pp. 6–11.
[38] K. Sycara, R. Glinton, B. Yu, J. Giampapa, S. Owens, M. Lewis, LTC C. Grindle, An
integrated approach to high-level information fusion, Inf. Fusion 10 (1) (2009)
25–50.
[39] S. Rodríguez, Y. De Paz, J. Bajo, J.M. Corchado, Social-based planning model for
multiagent systems, Expert Syst. Appl. 38 (10) (2011) 13005–13023.
[40] N. Jennings, M. Wooldridge, Agent Technology: Foundations, Applications and
Markets, Springer, Germany, 1998.
[41] J. Ferber, O. Gutknecht, F. Michel, From agents to organizations: an
organizational view of multi-agent systems, in: Proceedings of AgentOriented Software Engineering VI, 2003, pp. 214–230.
[42] O. Boissier, B. Gateau, Normative multi-agent organizations: Modeling,
support and control, in: G. Boella, L.W.N. van der Torre, H. Verhagen (Eds.),
Normative Multiagent Systems, Dagstuhl Seminar Proceedings 07122, Vol. II,
Germany, 2007. ISSN: 1862–4405.
[43] V. Dignum, A model for organizational interaction: based on agents, founded
in logic, PhD. Thesis, 2004.
[44] J. Pavon, J.J. Gomez-Sanz, Agent oriented software engineering with ingenias,
in: V. Marik, J.P. Müller, M. Pechoucek (Eds.), Proceedings of CEECMAS,
Heidelberg, Germany, 2003, pp. 394–403.
[45] J.F. Hubner, J.S. Sichman, O. Boissier, Using the Moise+ for a cooperative
framework of mass reorganization, in: Proceedings of the 17th Brazilian
Symposium on Artificial Intelligence – SBIA’04, 2004, pp. 506–515.
[46] F. Zambonelli, Abstractions and infrastructures for the design and
development of mobile agent organizations, in: Proceedings of the AgentOriented Software Engineering II, 2002, pp. 245–262.
[47] M. Esteva, Electronic Institutions: from specification to development,
Technical University of Catalonia, Spain, 2003 (Ph.D. Thesis).
[48] H.V.D. Parunak, J. Odell, Representing Social Structures in UML, in:
Wooldridge, M.J., Weiß, G., Ciancarini P. (Eds.), Agent-Oriented Software Eng.
II. Lecture Notes in Computer Science, vol. 2222, 2002, pp. 1–16. ISBN: 978-3540-43282-1.
[49] D.L. Hall, J. Llinas, An Introduction to multisensor information fusion, Proc.
IEEE 85 (1) (1997) 6–23.
[50] J. Bajo, J. Vicente, J.M. Corchado, C. Carrascosa, Y.D. Paz, V. Botti, J.F.D. Paz, An
execution time planner for the ARTIS agent architecture, Eng. Appl. Artif. Intell.
21 (2008) 769–784.
[51] S. Pang, T. Ban, Y. Kadobayashi, N. Kasabov, Personalized mode transductive
spanning SVM classification tree, Inf. Sci. 181 (2011) 2071–2085.
[52] R.S. Alonso, D.I. Tapia, J. Bajo, Ó. García, J.F. de Paz, J.M. Corchado,
Implementing a hardware-embedded reactive agents platform based on a
service-oriented architecture over heterogeneous wireless sensor networks,
Ad Hoc Netw. 11 (1) (2013) 151–166.
[53] J.F. de Paz, D.I. Tapia, R.S. Alonso, C.I. Pinzón, J. Bajo, J.M. Corchado, Mitigation
of the ground reflection effect in real-time locating systems based on wireless
sensor networks by using artificial neural networks, Int. J. Knowledge Inf. Syst.
34 (1) (2013) 193–217.
[54] D.I. Tapia, J.A. Fraile, S. Rodríguez, R.S. Alonso, J.M. Corchado, Integrating
hardware agents into an enhanced multi-agent architecture for Ambient
Intelligence systems, Inf. Sci. 222 (10) (2013) 47–65.
[55] C. Zato, A. Sanchez, G. Villarrubia, S. Rodriguez, J.M. Corchado, J. Bajo, Platform
for building large-scale agent-based systems, in: IEEE Conference on Evolving
and Adaptive Intelligent Systems, Madrid, Spain, 2012, pp. 69–73.
[56] D.I. Tapia, O. García, R.S. Alonso, F. Guevara, J. Catalina, R.A. Bravo, J.M.
Corchado, The n-Core Polaris Real-Time Locating System at the EvAAL
Competition, in: Evaluating
AAL
Systems Through Competitive
Benchmarking, Indoor Localization and Tracking Communications in
Computer and Information Science, 2012, pp. 92–106.
[57] D.I. Tapia, F. De la Prieta, S. Rodríguez, J. Bajo, J. M. Corchado, Organizations of
Agents in Information Fusion Environments, in: L. Antunes, H. S. Pinto (Eds.),
Progress in Artificial Intelligence – 15th Portuguese Conference on Artificial
Intelligence – EPIA 2011, Lisbon, Portugal, 2011, pp. 59–70.
[58] J.M. Corchado, J. Bajo, D.I. Tapia, A. Abraham, Using heterogeneous wireless
sensor networks in a telemonitoring system for healthcare, in: IEEE
Transactions on Information Technology in Biomedicine – Special Issue:
Affective and Pervasive Computing for Healthcare, 2009, pp. 663–670.
[59] J.F. De Paz, D.I. Tapia, R.S. Alonso, C.I. Pinzón, J. Bajo, J.M. Corchado, Mitigation
of the ground reflection effect in real-time locating systems based on wireless
sensor networks by using artificial neural networks, Knowl. Inf. Syst. 34 (2013)
193–217.
S. Rodríguez et al. / Information Fusion 23 (2015) 43–57
[60] R.R. Bouckaert, Bayesian Belief Networks: from Construction to Inference,
Utrecht, Netherlands, 1995.
[61] T. Verma, J. Pearl, An algorithm for deciding if a set of observed independencies
has a causal explanation, in: Proceedings of the Eighth Conference on
Uncertainty in Artificial Intelligence, 1992, pp. 323–330.
[62] N. Friedman, D. Geiger, M. Goldszmidt, Bayesian Network Classifiers, Machine
Learning 29 (1997) 131–163.
57
[63] ZigBee Standards Organization, ZigBee Specification Document 053474r13,
2006.
[64] Drools. The Business Logic Integration Platform <http://www.jboss.org/drools/
> (accessed December 2013).
[65] B. Berstel, Extending the RETE algorithm for event management, in: Temporal
Representation and Reasoning – Proceedings – Ninth International
Symposium, 2002, pp. 49–51.